xref: /linux/net/vmw_vsock/virtio_transport.c (revision b58b13f156c00c2457035b7071eaaac105fe6836)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * virtio transport for vsock
4  *
5  * Copyright (C) 2013-2015 Red Hat, Inc.
6  * Author: Asias He <asias@redhat.com>
7  *         Stefan Hajnoczi <stefanha@redhat.com>
8  *
9  * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s
10  * early virtio-vsock proof-of-concept bits.
11  */
12 #include <linux/spinlock.h>
13 #include <linux/module.h>
14 #include <linux/list.h>
15 #include <linux/atomic.h>
16 #include <linux/virtio.h>
17 #include <linux/virtio_ids.h>
18 #include <linux/virtio_config.h>
19 #include <linux/virtio_vsock.h>
20 #include <net/sock.h>
21 #include <linux/mutex.h>
22 #include <net/af_vsock.h>
23 
24 static struct workqueue_struct *virtio_vsock_workqueue;
25 static struct virtio_vsock __rcu *the_virtio_vsock;
26 static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */
27 static struct virtio_transport virtio_transport; /* forward declaration */
28 
29 struct virtio_vsock {
30 	struct virtio_device *vdev;
31 	struct virtqueue *vqs[VSOCK_VQ_MAX];
32 
33 	/* Virtqueue processing is deferred to a workqueue */
34 	struct work_struct tx_work;
35 	struct work_struct rx_work;
36 	struct work_struct event_work;
37 
38 	/* The following fields are protected by tx_lock.  vqs[VSOCK_VQ_TX]
39 	 * must be accessed with tx_lock held.
40 	 */
41 	struct mutex tx_lock;
42 	bool tx_run;
43 
44 	struct work_struct send_pkt_work;
45 	struct sk_buff_head send_pkt_queue;
46 
47 	atomic_t queued_replies;
48 
49 	/* The following fields are protected by rx_lock.  vqs[VSOCK_VQ_RX]
50 	 * must be accessed with rx_lock held.
51 	 */
52 	struct mutex rx_lock;
53 	bool rx_run;
54 	int rx_buf_nr;
55 	int rx_buf_max_nr;
56 
57 	/* The following fields are protected by event_lock.
58 	 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
59 	 */
60 	struct mutex event_lock;
61 	bool event_run;
62 	struct virtio_vsock_event event_list[8];
63 
64 	u32 guest_cid;
65 	bool seqpacket_allow;
66 
67 	/* These fields are used only in tx path in function
68 	 * 'virtio_transport_send_pkt_work()', so to save
69 	 * stack space in it, place both of them here. Each
70 	 * pointer from 'out_sgs' points to the corresponding
71 	 * element in 'out_bufs' - this is initialized in
72 	 * 'virtio_vsock_probe()'. Both fields are protected
73 	 * by 'tx_lock'. +1 is needed for packet header.
74 	 */
75 	struct scatterlist *out_sgs[MAX_SKB_FRAGS + 1];
76 	struct scatterlist out_bufs[MAX_SKB_FRAGS + 1];
77 };
78 
79 static u32 virtio_transport_get_local_cid(void)
80 {
81 	struct virtio_vsock *vsock;
82 	u32 ret;
83 
84 	rcu_read_lock();
85 	vsock = rcu_dereference(the_virtio_vsock);
86 	if (!vsock) {
87 		ret = VMADDR_CID_ANY;
88 		goto out_rcu;
89 	}
90 
91 	ret = vsock->guest_cid;
92 out_rcu:
93 	rcu_read_unlock();
94 	return ret;
95 }
96 
97 static void
98 virtio_transport_send_pkt_work(struct work_struct *work)
99 {
100 	struct virtio_vsock *vsock =
101 		container_of(work, struct virtio_vsock, send_pkt_work);
102 	struct virtqueue *vq;
103 	bool added = false;
104 	bool restart_rx = false;
105 
106 	mutex_lock(&vsock->tx_lock);
107 
108 	if (!vsock->tx_run)
109 		goto out;
110 
111 	vq = vsock->vqs[VSOCK_VQ_TX];
112 
113 	for (;;) {
114 		int ret, in_sg = 0, out_sg = 0;
115 		struct scatterlist **sgs;
116 		struct sk_buff *skb;
117 		bool reply;
118 
119 		skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
120 		if (!skb)
121 			break;
122 
123 		virtio_transport_deliver_tap_pkt(skb);
124 		reply = virtio_vsock_skb_reply(skb);
125 		sgs = vsock->out_sgs;
126 		sg_init_one(sgs[out_sg], virtio_vsock_hdr(skb),
127 			    sizeof(*virtio_vsock_hdr(skb)));
128 		out_sg++;
129 
130 		if (!skb_is_nonlinear(skb)) {
131 			if (skb->len > 0) {
132 				sg_init_one(sgs[out_sg], skb->data, skb->len);
133 				out_sg++;
134 			}
135 		} else {
136 			struct skb_shared_info *si;
137 			int i;
138 
139 			/* If skb is nonlinear, then its buffer must contain
140 			 * only header and nothing more. Data is stored in
141 			 * the fragged part.
142 			 */
143 			WARN_ON_ONCE(skb_headroom(skb) != sizeof(*virtio_vsock_hdr(skb)));
144 
145 			si = skb_shinfo(skb);
146 
147 			for (i = 0; i < si->nr_frags; i++) {
148 				skb_frag_t *skb_frag = &si->frags[i];
149 				void *va;
150 
151 				/* We will use 'page_to_virt()' for the userspace page
152 				 * here, because virtio or dma-mapping layers will call
153 				 * 'virt_to_phys()' later to fill the buffer descriptor.
154 				 * We don't touch memory at "virtual" address of this page.
155 				 */
156 				va = page_to_virt(skb_frag_page(skb_frag));
157 				sg_init_one(sgs[out_sg],
158 					    va + skb_frag_off(skb_frag),
159 					    skb_frag_size(skb_frag));
160 				out_sg++;
161 			}
162 		}
163 
164 		ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, skb, GFP_KERNEL);
165 		/* Usually this means that there is no more space available in
166 		 * the vq
167 		 */
168 		if (ret < 0) {
169 			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
170 			break;
171 		}
172 
173 		if (reply) {
174 			struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
175 			int val;
176 
177 			val = atomic_dec_return(&vsock->queued_replies);
178 
179 			/* Do we now have resources to resume rx processing? */
180 			if (val + 1 == virtqueue_get_vring_size(rx_vq))
181 				restart_rx = true;
182 		}
183 
184 		added = true;
185 	}
186 
187 	if (added)
188 		virtqueue_kick(vq);
189 
190 out:
191 	mutex_unlock(&vsock->tx_lock);
192 
193 	if (restart_rx)
194 		queue_work(virtio_vsock_workqueue, &vsock->rx_work);
195 }
196 
197 static int
198 virtio_transport_send_pkt(struct sk_buff *skb)
199 {
200 	struct virtio_vsock_hdr *hdr;
201 	struct virtio_vsock *vsock;
202 	int len = skb->len;
203 
204 	hdr = virtio_vsock_hdr(skb);
205 
206 	rcu_read_lock();
207 	vsock = rcu_dereference(the_virtio_vsock);
208 	if (!vsock) {
209 		kfree_skb(skb);
210 		len = -ENODEV;
211 		goto out_rcu;
212 	}
213 
214 	if (le64_to_cpu(hdr->dst_cid) == vsock->guest_cid) {
215 		kfree_skb(skb);
216 		len = -ENODEV;
217 		goto out_rcu;
218 	}
219 
220 	if (virtio_vsock_skb_reply(skb))
221 		atomic_inc(&vsock->queued_replies);
222 
223 	virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
224 	queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
225 
226 out_rcu:
227 	rcu_read_unlock();
228 	return len;
229 }
230 
231 static int
232 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
233 {
234 	struct virtio_vsock *vsock;
235 	int cnt = 0, ret;
236 
237 	rcu_read_lock();
238 	vsock = rcu_dereference(the_virtio_vsock);
239 	if (!vsock) {
240 		ret = -ENODEV;
241 		goto out_rcu;
242 	}
243 
244 	cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
245 
246 	if (cnt) {
247 		struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
248 		int new_cnt;
249 
250 		new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
251 		if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
252 		    new_cnt < virtqueue_get_vring_size(rx_vq))
253 			queue_work(virtio_vsock_workqueue, &vsock->rx_work);
254 	}
255 
256 	ret = 0;
257 
258 out_rcu:
259 	rcu_read_unlock();
260 	return ret;
261 }
262 
263 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
264 {
265 	int total_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM;
266 	struct scatterlist pkt, *p;
267 	struct virtqueue *vq;
268 	struct sk_buff *skb;
269 	int ret;
270 
271 	vq = vsock->vqs[VSOCK_VQ_RX];
272 
273 	do {
274 		skb = virtio_vsock_alloc_skb(total_len, GFP_KERNEL);
275 		if (!skb)
276 			break;
277 
278 		memset(skb->head, 0, VIRTIO_VSOCK_SKB_HEADROOM);
279 		sg_init_one(&pkt, virtio_vsock_hdr(skb), total_len);
280 		p = &pkt;
281 		ret = virtqueue_add_sgs(vq, &p, 0, 1, skb, GFP_KERNEL);
282 		if (ret < 0) {
283 			kfree_skb(skb);
284 			break;
285 		}
286 
287 		vsock->rx_buf_nr++;
288 	} while (vq->num_free);
289 	if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
290 		vsock->rx_buf_max_nr = vsock->rx_buf_nr;
291 	virtqueue_kick(vq);
292 }
293 
294 static void virtio_transport_tx_work(struct work_struct *work)
295 {
296 	struct virtio_vsock *vsock =
297 		container_of(work, struct virtio_vsock, tx_work);
298 	struct virtqueue *vq;
299 	bool added = false;
300 
301 	vq = vsock->vqs[VSOCK_VQ_TX];
302 	mutex_lock(&vsock->tx_lock);
303 
304 	if (!vsock->tx_run)
305 		goto out;
306 
307 	do {
308 		struct sk_buff *skb;
309 		unsigned int len;
310 
311 		virtqueue_disable_cb(vq);
312 		while ((skb = virtqueue_get_buf(vq, &len)) != NULL) {
313 			consume_skb(skb);
314 			added = true;
315 		}
316 	} while (!virtqueue_enable_cb(vq));
317 
318 out:
319 	mutex_unlock(&vsock->tx_lock);
320 
321 	if (added)
322 		queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
323 }
324 
325 /* Is there space left for replies to rx packets? */
326 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
327 {
328 	struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
329 	int val;
330 
331 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
332 	val = atomic_read(&vsock->queued_replies);
333 
334 	return val < virtqueue_get_vring_size(vq);
335 }
336 
337 /* event_lock must be held */
338 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
339 				       struct virtio_vsock_event *event)
340 {
341 	struct scatterlist sg;
342 	struct virtqueue *vq;
343 
344 	vq = vsock->vqs[VSOCK_VQ_EVENT];
345 
346 	sg_init_one(&sg, event, sizeof(*event));
347 
348 	return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
349 }
350 
351 /* event_lock must be held */
352 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
353 {
354 	size_t i;
355 
356 	for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
357 		struct virtio_vsock_event *event = &vsock->event_list[i];
358 
359 		virtio_vsock_event_fill_one(vsock, event);
360 	}
361 
362 	virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
363 }
364 
365 static void virtio_vsock_reset_sock(struct sock *sk)
366 {
367 	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
368 	 * under vsock_table_lock so the sock cannot disappear while we're
369 	 * executing.
370 	 */
371 
372 	sk->sk_state = TCP_CLOSE;
373 	sk->sk_err = ECONNRESET;
374 	sk_error_report(sk);
375 }
376 
377 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
378 {
379 	struct virtio_device *vdev = vsock->vdev;
380 	__le64 guest_cid;
381 
382 	vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
383 			  &guest_cid, sizeof(guest_cid));
384 	vsock->guest_cid = le64_to_cpu(guest_cid);
385 }
386 
387 /* event_lock must be held */
388 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
389 				      struct virtio_vsock_event *event)
390 {
391 	switch (le32_to_cpu(event->id)) {
392 	case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
393 		virtio_vsock_update_guest_cid(vsock);
394 		vsock_for_each_connected_socket(&virtio_transport.transport,
395 						virtio_vsock_reset_sock);
396 		break;
397 	}
398 }
399 
400 static void virtio_transport_event_work(struct work_struct *work)
401 {
402 	struct virtio_vsock *vsock =
403 		container_of(work, struct virtio_vsock, event_work);
404 	struct virtqueue *vq;
405 
406 	vq = vsock->vqs[VSOCK_VQ_EVENT];
407 
408 	mutex_lock(&vsock->event_lock);
409 
410 	if (!vsock->event_run)
411 		goto out;
412 
413 	do {
414 		struct virtio_vsock_event *event;
415 		unsigned int len;
416 
417 		virtqueue_disable_cb(vq);
418 		while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
419 			if (len == sizeof(*event))
420 				virtio_vsock_event_handle(vsock, event);
421 
422 			virtio_vsock_event_fill_one(vsock, event);
423 		}
424 	} while (!virtqueue_enable_cb(vq));
425 
426 	virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
427 out:
428 	mutex_unlock(&vsock->event_lock);
429 }
430 
431 static void virtio_vsock_event_done(struct virtqueue *vq)
432 {
433 	struct virtio_vsock *vsock = vq->vdev->priv;
434 
435 	if (!vsock)
436 		return;
437 	queue_work(virtio_vsock_workqueue, &vsock->event_work);
438 }
439 
440 static void virtio_vsock_tx_done(struct virtqueue *vq)
441 {
442 	struct virtio_vsock *vsock = vq->vdev->priv;
443 
444 	if (!vsock)
445 		return;
446 	queue_work(virtio_vsock_workqueue, &vsock->tx_work);
447 }
448 
449 static void virtio_vsock_rx_done(struct virtqueue *vq)
450 {
451 	struct virtio_vsock *vsock = vq->vdev->priv;
452 
453 	if (!vsock)
454 		return;
455 	queue_work(virtio_vsock_workqueue, &vsock->rx_work);
456 }
457 
458 static bool virtio_transport_can_msgzerocopy(int bufs_num)
459 {
460 	struct virtio_vsock *vsock;
461 	bool res = false;
462 
463 	rcu_read_lock();
464 
465 	vsock = rcu_dereference(the_virtio_vsock);
466 	if (vsock) {
467 		struct virtqueue *vq = vsock->vqs[VSOCK_VQ_TX];
468 
469 		/* Check that tx queue is large enough to keep whole
470 		 * data to send. This is needed, because when there is
471 		 * not enough free space in the queue, current skb to
472 		 * send will be reinserted to the head of tx list of
473 		 * the socket to retry transmission later, so if skb
474 		 * is bigger than whole queue, it will be reinserted
475 		 * again and again, thus blocking other skbs to be sent.
476 		 * Each page of the user provided buffer will be added
477 		 * as a single buffer to the tx virtqueue, so compare
478 		 * number of pages against maximum capacity of the queue.
479 		 */
480 		if (bufs_num <= vq->num_max)
481 			res = true;
482 	}
483 
484 	rcu_read_unlock();
485 
486 	return res;
487 }
488 
489 static bool virtio_transport_msgzerocopy_allow(void)
490 {
491 	return true;
492 }
493 
494 static bool virtio_transport_seqpacket_allow(u32 remote_cid);
495 
496 static struct virtio_transport virtio_transport = {
497 	.transport = {
498 		.module                   = THIS_MODULE,
499 
500 		.get_local_cid            = virtio_transport_get_local_cid,
501 
502 		.init                     = virtio_transport_do_socket_init,
503 		.destruct                 = virtio_transport_destruct,
504 		.release                  = virtio_transport_release,
505 		.connect                  = virtio_transport_connect,
506 		.shutdown                 = virtio_transport_shutdown,
507 		.cancel_pkt               = virtio_transport_cancel_pkt,
508 
509 		.dgram_bind               = virtio_transport_dgram_bind,
510 		.dgram_dequeue            = virtio_transport_dgram_dequeue,
511 		.dgram_enqueue            = virtio_transport_dgram_enqueue,
512 		.dgram_allow              = virtio_transport_dgram_allow,
513 
514 		.stream_dequeue           = virtio_transport_stream_dequeue,
515 		.stream_enqueue           = virtio_transport_stream_enqueue,
516 		.stream_has_data          = virtio_transport_stream_has_data,
517 		.stream_has_space         = virtio_transport_stream_has_space,
518 		.stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
519 		.stream_is_active         = virtio_transport_stream_is_active,
520 		.stream_allow             = virtio_transport_stream_allow,
521 
522 		.seqpacket_dequeue        = virtio_transport_seqpacket_dequeue,
523 		.seqpacket_enqueue        = virtio_transport_seqpacket_enqueue,
524 		.seqpacket_allow          = virtio_transport_seqpacket_allow,
525 		.seqpacket_has_data       = virtio_transport_seqpacket_has_data,
526 
527 		.msgzerocopy_allow        = virtio_transport_msgzerocopy_allow,
528 
529 		.notify_poll_in           = virtio_transport_notify_poll_in,
530 		.notify_poll_out          = virtio_transport_notify_poll_out,
531 		.notify_recv_init         = virtio_transport_notify_recv_init,
532 		.notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
533 		.notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
534 		.notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
535 		.notify_send_init         = virtio_transport_notify_send_init,
536 		.notify_send_pre_block    = virtio_transport_notify_send_pre_block,
537 		.notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
538 		.notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
539 		.notify_buffer_size       = virtio_transport_notify_buffer_size,
540 		.notify_set_rcvlowat      = virtio_transport_notify_set_rcvlowat,
541 
542 		.read_skb = virtio_transport_read_skb,
543 	},
544 
545 	.send_pkt = virtio_transport_send_pkt,
546 	.can_msgzerocopy = virtio_transport_can_msgzerocopy,
547 };
548 
549 static bool virtio_transport_seqpacket_allow(u32 remote_cid)
550 {
551 	struct virtio_vsock *vsock;
552 	bool seqpacket_allow;
553 
554 	seqpacket_allow = false;
555 	rcu_read_lock();
556 	vsock = rcu_dereference(the_virtio_vsock);
557 	if (vsock)
558 		seqpacket_allow = vsock->seqpacket_allow;
559 	rcu_read_unlock();
560 
561 	return seqpacket_allow;
562 }
563 
564 static void virtio_transport_rx_work(struct work_struct *work)
565 {
566 	struct virtio_vsock *vsock =
567 		container_of(work, struct virtio_vsock, rx_work);
568 	struct virtqueue *vq;
569 
570 	vq = vsock->vqs[VSOCK_VQ_RX];
571 
572 	mutex_lock(&vsock->rx_lock);
573 
574 	if (!vsock->rx_run)
575 		goto out;
576 
577 	do {
578 		virtqueue_disable_cb(vq);
579 		for (;;) {
580 			struct sk_buff *skb;
581 			unsigned int len;
582 
583 			if (!virtio_transport_more_replies(vsock)) {
584 				/* Stop rx until the device processes already
585 				 * pending replies.  Leave rx virtqueue
586 				 * callbacks disabled.
587 				 */
588 				goto out;
589 			}
590 
591 			skb = virtqueue_get_buf(vq, &len);
592 			if (!skb)
593 				break;
594 
595 			vsock->rx_buf_nr--;
596 
597 			/* Drop short/long packets */
598 			if (unlikely(len < sizeof(struct virtio_vsock_hdr) ||
599 				     len > virtio_vsock_skb_len(skb))) {
600 				kfree_skb(skb);
601 				continue;
602 			}
603 
604 			virtio_vsock_skb_rx_put(skb);
605 			virtio_transport_deliver_tap_pkt(skb);
606 			virtio_transport_recv_pkt(&virtio_transport, skb);
607 		}
608 	} while (!virtqueue_enable_cb(vq));
609 
610 out:
611 	if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
612 		virtio_vsock_rx_fill(vsock);
613 	mutex_unlock(&vsock->rx_lock);
614 }
615 
616 static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
617 {
618 	struct virtio_device *vdev = vsock->vdev;
619 	static const char * const names[] = {
620 		"rx",
621 		"tx",
622 		"event",
623 	};
624 	vq_callback_t *callbacks[] = {
625 		virtio_vsock_rx_done,
626 		virtio_vsock_tx_done,
627 		virtio_vsock_event_done,
628 	};
629 	int ret;
630 
631 	ret = virtio_find_vqs(vdev, VSOCK_VQ_MAX, vsock->vqs, callbacks, names,
632 			      NULL);
633 	if (ret < 0)
634 		return ret;
635 
636 	virtio_vsock_update_guest_cid(vsock);
637 
638 	virtio_device_ready(vdev);
639 
640 	return 0;
641 }
642 
643 static void virtio_vsock_vqs_start(struct virtio_vsock *vsock)
644 {
645 	mutex_lock(&vsock->tx_lock);
646 	vsock->tx_run = true;
647 	mutex_unlock(&vsock->tx_lock);
648 
649 	mutex_lock(&vsock->rx_lock);
650 	virtio_vsock_rx_fill(vsock);
651 	vsock->rx_run = true;
652 	mutex_unlock(&vsock->rx_lock);
653 
654 	mutex_lock(&vsock->event_lock);
655 	virtio_vsock_event_fill(vsock);
656 	vsock->event_run = true;
657 	mutex_unlock(&vsock->event_lock);
658 
659 	/* virtio_transport_send_pkt() can queue packets once
660 	 * the_virtio_vsock is set, but they won't be processed until
661 	 * vsock->tx_run is set to true. We queue vsock->send_pkt_work
662 	 * when initialization finishes to send those packets queued
663 	 * earlier.
664 	 * We don't need to queue the other workers (rx, event) because
665 	 * as long as we don't fill the queues with empty buffers, the
666 	 * host can't send us any notification.
667 	 */
668 	queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
669 }
670 
671 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
672 {
673 	struct virtio_device *vdev = vsock->vdev;
674 	struct sk_buff *skb;
675 
676 	/* Reset all connected sockets when the VQs disappear */
677 	vsock_for_each_connected_socket(&virtio_transport.transport,
678 					virtio_vsock_reset_sock);
679 
680 	/* Stop all work handlers to make sure no one is accessing the device,
681 	 * so we can safely call virtio_reset_device().
682 	 */
683 	mutex_lock(&vsock->rx_lock);
684 	vsock->rx_run = false;
685 	mutex_unlock(&vsock->rx_lock);
686 
687 	mutex_lock(&vsock->tx_lock);
688 	vsock->tx_run = false;
689 	mutex_unlock(&vsock->tx_lock);
690 
691 	mutex_lock(&vsock->event_lock);
692 	vsock->event_run = false;
693 	mutex_unlock(&vsock->event_lock);
694 
695 	/* Flush all device writes and interrupts, device will not use any
696 	 * more buffers.
697 	 */
698 	virtio_reset_device(vdev);
699 
700 	mutex_lock(&vsock->rx_lock);
701 	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
702 		kfree_skb(skb);
703 	mutex_unlock(&vsock->rx_lock);
704 
705 	mutex_lock(&vsock->tx_lock);
706 	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
707 		kfree_skb(skb);
708 	mutex_unlock(&vsock->tx_lock);
709 
710 	virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
711 
712 	/* Delete virtqueues and flush outstanding callbacks if any */
713 	vdev->config->del_vqs(vdev);
714 }
715 
716 static int virtio_vsock_probe(struct virtio_device *vdev)
717 {
718 	struct virtio_vsock *vsock = NULL;
719 	int ret;
720 	int i;
721 
722 	ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
723 	if (ret)
724 		return ret;
725 
726 	/* Only one virtio-vsock device per guest is supported */
727 	if (rcu_dereference_protected(the_virtio_vsock,
728 				lockdep_is_held(&the_virtio_vsock_mutex))) {
729 		ret = -EBUSY;
730 		goto out;
731 	}
732 
733 	vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
734 	if (!vsock) {
735 		ret = -ENOMEM;
736 		goto out;
737 	}
738 
739 	vsock->vdev = vdev;
740 
741 	vsock->rx_buf_nr = 0;
742 	vsock->rx_buf_max_nr = 0;
743 	atomic_set(&vsock->queued_replies, 0);
744 
745 	mutex_init(&vsock->tx_lock);
746 	mutex_init(&vsock->rx_lock);
747 	mutex_init(&vsock->event_lock);
748 	skb_queue_head_init(&vsock->send_pkt_queue);
749 	INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
750 	INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
751 	INIT_WORK(&vsock->event_work, virtio_transport_event_work);
752 	INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
753 
754 	if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET))
755 		vsock->seqpacket_allow = true;
756 
757 	vdev->priv = vsock;
758 
759 	ret = virtio_vsock_vqs_init(vsock);
760 	if (ret < 0)
761 		goto out;
762 
763 	for (i = 0; i < ARRAY_SIZE(vsock->out_sgs); i++)
764 		vsock->out_sgs[i] = &vsock->out_bufs[i];
765 
766 	rcu_assign_pointer(the_virtio_vsock, vsock);
767 	virtio_vsock_vqs_start(vsock);
768 
769 	mutex_unlock(&the_virtio_vsock_mutex);
770 
771 	return 0;
772 
773 out:
774 	kfree(vsock);
775 	mutex_unlock(&the_virtio_vsock_mutex);
776 	return ret;
777 }
778 
779 static void virtio_vsock_remove(struct virtio_device *vdev)
780 {
781 	struct virtio_vsock *vsock = vdev->priv;
782 
783 	mutex_lock(&the_virtio_vsock_mutex);
784 
785 	vdev->priv = NULL;
786 	rcu_assign_pointer(the_virtio_vsock, NULL);
787 	synchronize_rcu();
788 
789 	virtio_vsock_vqs_del(vsock);
790 
791 	/* Other works can be queued before 'config->del_vqs()', so we flush
792 	 * all works before to free the vsock object to avoid use after free.
793 	 */
794 	flush_work(&vsock->rx_work);
795 	flush_work(&vsock->tx_work);
796 	flush_work(&vsock->event_work);
797 	flush_work(&vsock->send_pkt_work);
798 
799 	mutex_unlock(&the_virtio_vsock_mutex);
800 
801 	kfree(vsock);
802 }
803 
804 #ifdef CONFIG_PM_SLEEP
805 static int virtio_vsock_freeze(struct virtio_device *vdev)
806 {
807 	struct virtio_vsock *vsock = vdev->priv;
808 
809 	mutex_lock(&the_virtio_vsock_mutex);
810 
811 	rcu_assign_pointer(the_virtio_vsock, NULL);
812 	synchronize_rcu();
813 
814 	virtio_vsock_vqs_del(vsock);
815 
816 	mutex_unlock(&the_virtio_vsock_mutex);
817 
818 	return 0;
819 }
820 
821 static int virtio_vsock_restore(struct virtio_device *vdev)
822 {
823 	struct virtio_vsock *vsock = vdev->priv;
824 	int ret;
825 
826 	mutex_lock(&the_virtio_vsock_mutex);
827 
828 	/* Only one virtio-vsock device per guest is supported */
829 	if (rcu_dereference_protected(the_virtio_vsock,
830 				lockdep_is_held(&the_virtio_vsock_mutex))) {
831 		ret = -EBUSY;
832 		goto out;
833 	}
834 
835 	ret = virtio_vsock_vqs_init(vsock);
836 	if (ret < 0)
837 		goto out;
838 
839 	rcu_assign_pointer(the_virtio_vsock, vsock);
840 	virtio_vsock_vqs_start(vsock);
841 
842 out:
843 	mutex_unlock(&the_virtio_vsock_mutex);
844 	return ret;
845 }
846 #endif /* CONFIG_PM_SLEEP */
847 
848 static struct virtio_device_id id_table[] = {
849 	{ VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
850 	{ 0 },
851 };
852 
853 static unsigned int features[] = {
854 	VIRTIO_VSOCK_F_SEQPACKET
855 };
856 
857 static struct virtio_driver virtio_vsock_driver = {
858 	.feature_table = features,
859 	.feature_table_size = ARRAY_SIZE(features),
860 	.driver.name = KBUILD_MODNAME,
861 	.driver.owner = THIS_MODULE,
862 	.id_table = id_table,
863 	.probe = virtio_vsock_probe,
864 	.remove = virtio_vsock_remove,
865 #ifdef CONFIG_PM_SLEEP
866 	.freeze = virtio_vsock_freeze,
867 	.restore = virtio_vsock_restore,
868 #endif
869 };
870 
871 static int __init virtio_vsock_init(void)
872 {
873 	int ret;
874 
875 	virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
876 	if (!virtio_vsock_workqueue)
877 		return -ENOMEM;
878 
879 	ret = vsock_core_register(&virtio_transport.transport,
880 				  VSOCK_TRANSPORT_F_G2H);
881 	if (ret)
882 		goto out_wq;
883 
884 	ret = register_virtio_driver(&virtio_vsock_driver);
885 	if (ret)
886 		goto out_vci;
887 
888 	return 0;
889 
890 out_vci:
891 	vsock_core_unregister(&virtio_transport.transport);
892 out_wq:
893 	destroy_workqueue(virtio_vsock_workqueue);
894 	return ret;
895 }
896 
897 static void __exit virtio_vsock_exit(void)
898 {
899 	unregister_virtio_driver(&virtio_vsock_driver);
900 	vsock_core_unregister(&virtio_transport.transport);
901 	destroy_workqueue(virtio_vsock_workqueue);
902 }
903 
904 module_init(virtio_vsock_init);
905 module_exit(virtio_vsock_exit);
906 MODULE_LICENSE("GPL v2");
907 MODULE_AUTHOR("Asias He");
908 MODULE_DESCRIPTION("virtio transport for vsock");
909 MODULE_DEVICE_TABLE(virtio, id_table);
910